[SOLVED] MIMCAP Process Variation

Hello Forum,

Why is the process variation of MIMCAP worse than the process variation of a MOSCAP?

The paper linked here
says that MIMCAP can show about ±15% variation over process corners.

My own experiments in the 189nm TSMC process shows the variation to be 20% for a 10pF MIMCAP amd it increases from slow to fast corners.

Why would MIMCAP have any process variation since it is only METAL and SiO₂.
Shouldn't it be constant over process corners? What changes over corners to affect the MIMCAP?

I think this includes process (SiO₂ thickness) variation over several wafer lots. The variation on the same wafer will be much lower. VT variations should be insignificant.

Corners include the process variation, of course. Here, slow is assigned to min., fast to max. variation.

Higher interlayer thicknesses and metal layer dimensions are controlled with less accuracy than gate oxide thickness and diffusion layer dimensions.

erikl said:

Higher interlayer thicknesses and metal layer dimensions are controlled with less accuracy than gate oxide thickness and diffusion layer dimensions.

Click to expand...

But since the higher layers are thicker, wouldn't the thickness be easier to control especially since the processing technology for that should be quite mature by now?
Or is it like that they just do not care for its thickness like for the gate oxide since it is not as critical.

nitishn5 said:

But since the higher layers are thicker, wouldn't the thickness be easier to control especially since the processing technology for that should be quite mature by now?

Click to expand...

The thickness - I think: yes. But what about the dimensions? Degree of under-etching, perhaps?

nitishn5 said:

Or is it like that they just do not care for its thickness like for the gate oxide since it is not as critical.

Click to expand...

Normally, this is true. But it shouldn't be so for a process for which they offer the MIMCAP option. May be they simply didn't care about the MIMCAP matching figure and simply stated a very "secure" value. :-(

One of the 180nm processes which I had used (other fab) gave essentially better matching values for their MIMCAPs: View attachment MIM-cap_mismatch.pdf

Thanks a lot erikl.